1. Distribution analysis
2. Correlation analysis
- 2.1. JIF and 5-Year JIF
- 2.2. AIS, Eigenfactor, Immediacy Index and Total Citations
3. Regression analysis
- 3.1 Information Science & Library Science model

This is the R Markdown Notebook created for the JCI analysis. Notebook and data are available at GitHub.

source('functions.R')
library(reldist)
library(gglorenz)
library(ggplot2)
library(GGally)
library(dplyr)
library(ggnewscale)

set.seed(13)

First line only can be executed if you have downloaded data from JCR in multiple files, otherwise, use the second line which includes the anonymized JCR data.

#df <- read_jcr('data/files/')
df <- read.delim('data/jcr.tsv', stringsAsFactors=FALSE, check.names=FALSE)

Some descriptive statistics of the dataset.

summary(df$`2020 JCI`)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's 
##  0.0000  0.3400  0.6400  0.8058  1.0300 77.6400      66

1. Distribution analysis

Gini Index is calculated for each JCR indicator.

gini(df$`2020 JIF`[which(!is.na(df$`2020 JIF`))])

## [1] 0.4444105

gini(df$`2020 JCI`[which(!is.na(df$`2020 JCI`))])

## [1] 0.4365864

gini(df$`5 Year JIF` [which(!is.na(df$`5 Year JIF`))])

## [1] 0.4403715

gini(df$`Immediacy Index`[which(!is.na(df$`Immediacy Index`))])

## [1] 0.6220866

gini(df$`Eigenfactor`[which(!is.na(df$`Eigenfactor`))])

## [1] 0.8019052

gini(df$`Total Citations`[which(!is.na(df$`Total Citations`))])

## [1] 0.7918109

gini(df$`Article Influence Score`[which(!is.na(df$`Article Influence Score`))])

## [1] 0.5499661

A new data.frame is created to obtain the Lorenz curve in an easy way.

df_den <- data.frame()

for(x in c('2020 JCI','2020 JIF', '5 Year JIF', 'Immediacy Index', 'Eigenfactor', 'Total Citations', 'Article Influence Score')){
  aux <-df[,c('Cat','DB',x)]
  names(aux)[3] <- 'Value'
  aux$Indicator <- x
  df_den <- rbind.data.frame(df_den, aux, stringsAsFactors = FALSE)
}

df_den[which(df_den$Indicator=='2020 JCI'),'Indicator'] <- 'Journal Citation Indicator'
df_den[which(df_den$Indicator=='2020 JIF'),'Indicator'] <- 'Journal Impact Factor'
df_den[which(df_den$Indicator=='5 Year JIF'),'Indicator']  <- '5-Year Journal Impact Factor'

The Lorenz curve is obtained.

ggplot(df_den, aes(x=Value, color=Indicator)) +
  stat_lorenz(desc=FALSE, alpha=0.7, size=1) +
  geom_abline(size=0.8) +
  scale_color_manual(values=c('#fb0007','#129176', '#e5c421', '#f46f08', '#4cacd0', '#291b37', '#de8bb7'))+
  theme_light() +
  theme(panel.grid=element_blank(),
        legend.title=element_blank(),
        text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=12),
        axis.ticks=element_line(color='black'),
        legend.position='bottom',
        panel.border=element_rect(colour='black'),
        strip.background=element_rect(colour='black', fill='black'),
        strip.text=element_text(size=14),
        legend.text=element_text(size=10))+
  guides(fill=guide_colorbar(label.position='bottom',
                               title.position='left', title.vjust=0.85)) +
  labs(y='Cummulative percentage of indicator', x='Cummulative percentage of journals')

2. Correlation analysis

A new data.frame is created for obtaining a matrix of plots.

m_plot <- df[,c('2020 JCI', '2020 JIF', '5 Year JIF', 'Immediacy Index', 'Eigenfactor', 'Total Citations', 'Article Influence Score', 'DB')]
names(m_plot)[c(1,2,7)] <- c('JCI', 'JIF', 'AIS')

The matrix of plots with the correlations and distribution of data is generated. All of them are significant.

ggpairs(m_plot, columns=1:7,
        upper=list(continuous=wrap('cor', size=5, method='pearson')),
        lower=list(continuous=wrap('points', alpha=0.6)))+
  theme(text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=11),
        axis.ticks=element_line(color='black'),
        strip.background=element_rect(colour='black', fill='black'),
        strip.text=element_text(size=12, color='white'),
        legend.position='none'
        )

Pearson correlations between the seven indicators are calculated.

df_cor <- data.frame()
for(x in c('2020 JIF', '5 Year JIF', 'Immediacy Index', 'Eigenfactor', 'Total Citations', 'Article Influence Score')){
  aux <-biblio_cor(df, x, '2020 JCI')
  aux$Indicator <- x
  df_cor <- rbind.data.frame(df_cor, aux, stringsAsFactors = FALSE)
}

df_cor <- df_cor[which(!is.na(df_cor$Correlation)),]

ESI field are assigned to each Web of Science Category.

esi <- read.csv2('data/final_map.csv', stringsAsFactors=FALSE)
esi[which(esi$WC=="Women's Studies"), 'WC'] <- 'Womens Studies'
esi$WC <- toupper(esi$WC)

df_cor <- inner_join(df_cor, esi[,c('WC', 'ESI')], by=c('Cat'='WC'))

2.1. JIF and 5-Year JIF

Firstly, correlations between JCI and JIF and 5-Year JIF are calculated.

jci_jif <- df_cor[which((df_cor$Indicator %in% c('2020 JIF', '5 Year JIF')) & (df_cor$DB %in% c('SSCI', 'SCIE'))),]

jci_jif[which(jci_jif$DB=='SSCI'),'DB'] <- 'Social Sciences'
jci_jif[which(jci_jif$DB=='SCIE'),'DB'] <- 'Science'

jci_jif[which(jci_jif$Indicator=='2020 JIF'),'Indicator'] <- 'Journal Impact Factor'
jci_jif[which(jci_jif$Indicator=='5 Year JIF'),'Indicator'] <- '5-Year Journal Impact Factor'

jci_jif$Indicator <- factor(jci_jif$Indicator, levels=c('Journal Impact Factor', '5-Year Journal Impact Factor'), ordered=TRUE)

There are 13 correlations with a p-value less than 0.01 (6 in JIF and 7 in 5-Year JIF).

dim(jci_jif[which(jci_jif$p > 0.001),])

## [1] 13  9

mean(jci_jif[which(jci_jif$p < 0.001 & jci_jif$Indicator=='Journal Impact Factor'),'Correlation'])

## [1] 0.8923508

mean(jci_jif[which(jci_jif$p < 0.001 & jci_jif$Indicator=='Journal Impact Factor' & jci_jif$DB == 'Science'),'Correlation'])

## [1] 0.9036724

mean(jci_jif[which(jci_jif$p < 0.001 & jci_jif$Indicator=='Journal Impact Factor' & jci_jif$DB == 'Social Sciences'),'Correlation'])

## [1] 0.8571731

Boxplot with correlations greater than 0.6 is generated.

ggplot(data=jci_jif[which(jci_jif$p < 0.001),], aes(y=Correlation, x=DB, group=DB))+
  geom_boxplot(outlier.shape=NA, width=0.8) +
  geom_jitter(data=jci_jif[which(jci_jif$Indicator=='Journal Impact Factor' & jci_jif$p < 0.001),], aes(fill=Correlation, size=Docs), shape=21, color='black', stroke=0.5, alpha=0.6)+
  scale_fill_gradient(low='white', high='#eb2701', na.value=NA, breaks=c(0.6, 0.8, 1), limits=c(0.6,1))+
  new_scale_fill()+
  geom_jitter(data=jci_jif[which(jci_jif$Indicator=='5-Year Journal Impact Factor' & jci_jif$p < 0.001),], aes(fill=Correlation, size=Docs), shape=21, color='black', stroke=0.5, alpha=0.6)+
  scale_fill_gradient(low='white', high='blue', na.value=NA, breaks=c(0.6, 0.8, 1), limits=c(0.6,1))+
  ylim(c(0.6,1))+
  scale_size_continuous(range=c(1,5), breaks=c(0, 25000,50000, 100000, 125000, 150000, 200000))+
  theme_light()+
  theme(panel.grid=element_blank(),
        text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=11),
        axis.ticks=element_line(color='black'),
        legend.position='none',
        legend.box='vertical',
        panel.border=element_rect(colour='black'),
        strip.background=element_rect(colour='black', fill='black'),
        strip.text = element_text(size=14))+
  guides(fill=guide_colorbar(label.position='bottom',
                             title.position='left', title.vjust=0.85)) +
  labs(x='', y='Correlation with JCI')+
  facet_wrap(~Indicator)

The complete boxplot does not vary much, only 7 more categories are included.

ggplot(data=jci_jif, aes(y=Correlation, x=DB, group=DB))+
  geom_boxplot(outlier.shape=NA, width=0.8) +
  geom_jitter(data=jci_jif[which(jci_jif$Indicator=='Journal Impact Factor'),], aes(fill=Correlation, size=Docs), shape=21, color='black', stroke=0.5, alpha=0.6)+
  scale_fill_gradient(low='white', high='#eb2701', na.value=NA, breaks=c(0.5, 0.75, 1), limits=c(0.5,1))+
  new_scale_fill()+
  geom_jitter(data=jci_jif[which(jci_jif$Indicator=='5-Year Journal Impact Factor'),], aes(fill=Correlation, size=Docs), shape=21, color='black', stroke=0.5, alpha=0.6)+
  scale_fill_gradient(low='white', high='blue', na.value=NA, breaks=c(0, 0.5, 1), limits=c(0,1))+
  scale_size_continuous(range=c(1,5), breaks=c(0, 25000,50000, 100000, 125000, 150000, 200000))+
  theme_light()+
  theme(panel.grid=element_blank(),
        text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=11),
        axis.ticks=element_line(color='black'),
        legend.position='none',
        legend.box='vertical',
        panel.border=element_rect(colour='black'),
        strip.background=element_rect(colour='black', fill='black'),
        strip.text = element_text(size=14))+
  guides(fill=guide_colorbar(label.position='bottom',
                             title.position='left', title.vjust=0.85)) +
  labs(x='')+
  facet_wrap(~Indicator)

2.2. AIS, Eigenfactor, Immediacy Index and Total Citations

The correlations between the JCI and the rest of the indicators are calculated, but taking into account only those categories with more than 25 journals. This is because a few small categories greatly alter the results.

ot_indicators <- df_cor[which((!(df_cor$Indicator %in% c('2020 JIF', '5 Year JIF'))) & df_cor$Journals>25),]

ot_indicators[which(ot_indicators$DB=='SSCI'),'DB'] <- 'Social\nSciences'
ot_indicators[which(ot_indicators$DB=='SCIE'),'DB'] <- 'Science'
ot_indicators[which(ot_indicators$DB=='AHCI'),'DB'] <- 'Arts &\nHumanities'
ot_indicators[which(ot_indicators$DB=='ESCI'),'DB'] <- 'Emerging\nSources'

There are 118 correlations with a p-value less than 0.01 (43 in Total Citations, 33 in Immediacy Index and Eigenfactor, and 9 in Article Influence Score).

table(ot_indicators[which(ot_indicators$p >= 0.01),'Indicator'])

## 
## Article Influence Score             Eigenfactor         Immediacy Index 
##                       9                      33                      33 
##         Total Citations 
##                      43

Boxplots are generated.

ggplot(data=ot_indicators, aes(y=Correlation, x=DB, group=DB))+
  geom_boxplot(outlier.shape=NA, width=0.8) +
  geom_jitter(aes(fill=Correlation, size=Docs), shape=21, color='black', stroke=0.5, alpha=0.6)+
  scale_fill_gradient(low='white', high='#eb2701', na.value=NA, breaks=c(0, 0.5, 1), limits=c(0,1))+
  scale_size_continuous(range=c(0.1,3.5), breaks=c(0, 25000,50000, 100000, 125000, 150000, 200000))+
  theme_light()+
  theme(panel.grid=element_blank(),
        text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=10.5),
        axis.ticks=element_line(color='black'),
        legend.position='bottom',
        panel.border=element_rect(colour='black'),
        strip.background=element_rect(colour='black', fill='black'),
        strip.text=element_text(size=14))+
  guides(fill = guide_colorbar(label.position='bottom',
                               title.position='left', title.vjust=0.85)) +
  labs(x='')+
  facet_wrap(~Indicator)

3. Regression analysis

Firstly, we rank JIF and JCI values by categories.

df_rank <- df[which(!(is.na(df$`2020 JIF`) | is.na(df$`2020 JCI`))),]

length(unique(df_rank$Category))

## [1] 236

df_rank$quantile_jif <- NA
df_rank$quantile_jci <- NA

for(ca in unique(df$Category)){
  df_rank[which(df_rank$Category==ca),'quantile_jif'] <- trunc(rank(-df_rank[which(df_rank$Category==ca),'2020 JIF'], ties.method = 'random'))
  df_rank[which(df_rank$Category==ca),'quantile_jci'] <- trunc(rank(-df_rank[which(df_rank$Category==ca),'2020 JCI'], ties.method = 'random'))
}

A linear regression model is generated with JIF and JCI ranks.

lm_g <- lm(quantile_jif~quantile_jci, df_rank[which(df_rank$DB %in% c('SSCI', 'SCIE')),])
summary(lm_g)

## 
## Call:
## lm(formula = quantile_jif ~ quantile_jci, data = df_rank[which(df_rank$DB %in% 
##     c("SSCI", "SCIE")), ])
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -206.610   -6.640   -1.233    6.139  211.112 
## 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  3.651088   0.213643   17.09   <2e-16 ***
## quantile_jci 0.947728   0.002255  420.36   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 20.43 on 20030 degrees of freedom
## Multiple R-squared:  0.8982, Adjusted R-squared:  0.8982 
## F-statistic: 1.767e+05 on 1 and 20030 DF,  p-value: < 2.2e-16

The models obtained for SSCI and SCIE journals have a good R².

lm_ssci <- lm(quantile_jif~quantile_jci, df_rank[which(df_rank$DB == 'SSCI'),])
summary(lm_ssci)$r.squared

## [1] 0.8851921

lm_scie <- lm(quantile_jif~quantile_jci, df_rank[which(df_rank$DB == 'SCIE'),])
summary(lm_scie)$r.squared

## [1] 0.9029314

df_r2 <- data.frame(quantile_jci=c(Inf,Inf),
                    quantile_jif=c(10,10),
                    text=c(paste0('y==', format(coef(lm_ssci)[1], digits = 2), '+', format(coef(lm_ssci)[2], digits = 2), '~x~', '~~italic(r)^2==', round(summary(lm_ssci)$r.squared,3),''),
                           paste0('y==', format(coef(lm_scie)[1], digits = 2), '+', format(coef(lm_scie)[2], digits = 2), '~x~', '~~italic(r)^2==', round(summary(lm_scie)$r.squared,3),'')),
                    DB=c('SSCI','SCIE'))

ggplot(data=df_rank[which(df_rank$DB %in% c('SSCI', 'SCIE')),], aes(x=quantile_jci, y=quantile_jif)) +
  geom_point(color='#1f77b4', size=1, alpha=0.75, stroke=0)+
  geom_smooth(formula = 'y ~ x', method='lm', color='black')+
  geom_text(data=df_r2, aes(x=quantile_jci, y=quantile_jif, label=text, group=DB),
            parse = TRUE, hjust=1.05) +
  theme_light() +
  labs(x='JCI rank', y='JIF rank') +
  facet_wrap(.~DB, scales = 'free') +
  theme(text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=11),
        axis.ticks=element_line(color='black'),
        panel.border=element_rect(color='black'),
        strip.background=element_rect(color='black', fill='black'),
        strip.text=element_text(size=12, color='white'),
        legend.position='none'
  )

The average R² value of the linear regression models is calculated.

df_r <- data.frame(category=unique(df_rank$Category[which(df_rank$DB %in% c('SSCI', 'SCIE'))]),
                   r2=NA,
                   DB=NA,
                   stringsAsFactors = FALSE)

for(ca in unique(df_r$category)){
  lm_aux <- NA
  lm_aux <-lm(quantile_jci~quantile_jif,df_rank[which(df_rank$Category==ca & !is.na(df_rank$`2020 JIF`) & !is.na(df_rank$`2020 JCI`)),])
  df_r[which(df_r$category==ca),'r2'] <- summary(lm_aux)$r.squared
  df_r[which(df_r$category==ca),'DB'] <- unique(df_rank[which(df_rank$Category==ca),'DB'])
}

mean(df_r$r2, na.rm = TRUE)

## [1] 0.8064401

mean(df_r$r2[which(df_r$DB=='SSCI')], na.rm = TRUE)

## [1] 0.7380394

mean(df_r$r2[which(df_r$DB=='SCIE')], na.rm = TRUE)

## [1] 0.8287279

3.1 Information Science & Library Science model

As specific study case, the linear regression model for Information Science & Library Science is generated.

df_lib <- df[which(df$Cat == 'INFORMATION SCIENCE & LIBRARY SCIENCE'),]
df_lib_train <- df_lib[which(!(is.na(df_lib$`2020 JIF`) | is.na(df_lib$`2020 JCI`))),]

df_lib_train$quantile_jif <- trunc(rank(-df_lib_train[,'2020 JIF'], ties.method = 'first'))
df_lib_train$quantile_jci <- trunc(rank(-df_lib_train[,'2020 JCI'], ties.method = 'first'))

lib_lm <- lm(quantile_jif~quantile_jci, df_lib_train)

JIF ranking is predicted and compared to the original JIF ranking.

df_lib_predict <- df_lib[which(!(is.na(df_lib$`2020 JCI`))),]
df_lib_predict$quantile_jci <- trunc(rank(-df_lib_predict[,'2020 JCI'], ties.method = 'first'))
df_lib_predict$quantile_jif <- trunc(rank(-df_lib_predict[,'2020 JIF'], ties.method = 'first'))
df_lib_predict$quantile_jif[which(is.na(df_lib_predict$`2020 JIF`))] <- NA

df_lib_predict$quantile_jci_p <- predict(lib_lm, newdata = df_lib_predict)
df_lib_predict$quantile_jci_p <- rank(as.numeric(df_lib_predict$quantile_jci_p), ties.method = 'random')

ggplot() +
  geom_point(data=df_lib_predict, aes(x='JIF', y=quantile_jif, color=DB), alpha=0.7) +
  geom_point(data=df_lib_predict, aes(x='Predicted JIF', y=quantile_jci_p, color=DB), alpha=0.7) +
  geom_segment(data =df_lib_predict, aes(x = 'Predicted JIF', xend = 'JIF', y = quantile_jci_p, yend = quantile_jif),
               inherit.aes = FALSE,
               color='grey20',
               size=0.25) +
  geom_text(aes(y=c(1,dim(df_lib_predict)[1]/4, 2*dim(df_lib_predict)[1]/4, 3*dim(df_lib_predict)[1]/4),
                x=Inf,
                label=c('Q1','Q2','Q3','Q4')),
            hjust=1.2, vjust=1)+
  scale_color_manual(values=c('#fc4e08', '#28b7c1'), labels = c('ESCI journals', 'SSCI journals'))+
  geom_hline(yintercept = c(dim(df_lib_predict)[1]/4, 2*dim(df_lib_predict)[1]/4, 3*dim(df_lib_predict)[1]/4), size=0.25)+
  coord_flip()+
  scale_x_discrete(limits=rev)+
  scale_y_continuous(trans = "reverse", breaks = c(1,50,100,150)) +
  labs(title = '', y='Ranking position', x='')+
  theme_classic()+
  theme(text=element_text(family='Arial', size=12.5, color='black'),
        axis.text=element_text(color='black', size=11),
        axis.ticks=element_line(color='black'),
        legend.position = 'bottom',
        legend.title = element_blank())

Is there a need for a new journal metric? Correlations between JCR Impact Factor metrics and the Journal Citation Indicator - JCI

Daniel Torres-Salinas, Pilar Valderrama-Baca, Wenceslao Arroyo-Machado